Color ultrasound image processing control system based on data feedback

Abstract

The invention discloses a color ultrasound image processing control system based on data feedback, and relates to the technical field of color ultrasound image processing. The color ultrasound image processing control system based on data feedback comprises an acquisition and generation module used for constructing an original data set, a scoring module, quantizing image quality based on SNR and MAR numerical values, a screening module, screening the constructed original data set based on an obtained screening threshold value, removing repeated data, and an execution module. And feeding back the processed image data and the context information to the color Doppler ultrasound image display equipment. According to the method, the original data set is constructed, SNR and MAR are combined, the clinical scene weight is introduced, the image recognition accuracy is improved, compared with single gray level evaluation, the method can better meet the actual requirements of different examination parts, then a neural network model is adopted to learn a dynamic threshold value from historical data, compared with a fixed threshold value, the redundant data compression rate is improved, and the image recognition accuracy is improved. And meanwhile, the effective information retention rate is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of color ultrasound image processing, in particular to a color ultrasound image processing control system based on data feedback. BACKGROUND

[0002] In the existing color ultrasound image processing technology, image quality evaluation relies on single gray scale features or manual experience, and fixed thresholds are used for repeated data screening, and there is a lack of deep integration with clinical scenarios, resulting in low processing efficiency and poor adaptability.

[0003] Although the prior art compares color ultrasound image data with a database to eliminate useless data and reconstruct the remaining data to meet display requirements, simply comparing with the database to eliminate data may be the data required by relevant personnel, and it is not possible to eliminate these data based on the demand of corresponding personnel, and feedback after eliminating these data will affect the analysis of the test results. Finally, the data feedback capability of the data feedback device is strong or weak, and some data feedback devices have remaining capacity in addition to the current image display, and unreasonable allocation of data feedback resources will result in low color ultrasound image feedback efficiency, inaccurate test result analysis and other consequences.

[0004] For example, a color ultrasound image processing control system based on data feedback disclosed in Chinese Patent No. CN114756542B analyzes the data feedback capability of different color ultrasound data feedback devices through big data, determines whether historical image data can be added or whether part of the current image data needs to be screened out based on the analysis result, if historical image data can be added, historical image data is retrieved through the creation of the shortest retrieval channel, which reduces the delay of data retrieval, and then compares the historical image data capacity with the feedback device capability to predict whether part of the data needs to be screened out.

[0005] The above scheme is only based on the pixel features of the image itself, such as gray scale variance and edge intensity, to calculate the quality score, without combining device state parameters and clinical scenarios, resulting in low matching degree between the score and actual diagnosis requirements, and by eliminating duplicate data, without considering patient medical history and test type, which is easy to cause effective information loss or excessive redundant data. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a color ultrasound image processing control system based on data feedback, which solves the problems raised in the background art.

[0007] To achieve the above purpose, the present application realizes the following technical scheme:

[0008] A color ultrasound image processing control system based on data feedback, comprising:

[0009] An acquisition and generation module is used to acquire color ultrasound image data, device state parameters, and construct an original data set based on structured patient history labels and examination types obtained from a hospital information system;

[0010] A scoring module is used to perform quality scoring based on the constructed original data set, and quantize image quality based on SNR and MAR values;

[0011] A screening module is used to find a screening threshold for repeated data from historical data sets based on a neural network model, and screen the constructed original data set based on the obtained screening threshold to remove repeated data;

[0012] An execution module is used to perform a screening strategy on the original data set based on feature extraction of a target observation area of the color ultrasound image, and feed the processed image data and context information to a color ultrasound image display device.

[0013] Further improvement of the technical scheme of the application is that in the acquisition and generation module, the device state parameters include gain, depth and frequency, the patient history labels include numerical features and category features, and the examination types include category vectors.

[0014] The color ultrasound image data is composed of a 2D gray matrix, which is represented as a one-dimensional vector by the acquisition and generation module, the device state parameters are composed of a group of numerical parameters of gain, depth and frequency, which are directly represented as vectors by the acquisition and generation module, the numerical features of the patient history labels include numerical components of age and disease duration, and the category features are gender and disease history, which are encoded into numerical values by the acquisition and generation module, and the acquisition and generation module combines all the above-mentioned features to form a composite feature vector, which is specifically:

[0015] The acquisition and generation module acquires the original color ultrasound matrix I i ∈R H×W , rearranges the matrix in row priority order into a column vector to obtain a flattened image vector , concatenates a device state parameter vector containing original device parameters , a patient history label vector containing original patient history labels , and an examination type vector containing sample examination types to establish a composite feature vector The original data set constructed by the acquisition and generation module includes a set of all composite feature vectors.

[0016] Further, the application vectorizes the color ultrasound image data, including: wherein, is the flattened image vector of the i-th sample, which belongs to R M, M is the dimension of the flattened vector, i i is the original color ultrasound image matrix of the i-th sample, belonging to R H×W , vec(·) is the matrix flattening operator, which rearranges the matrix in row-major order to a column vector, where M = H × W, H is the image height, W is the image width, and finally the image data is converted to vector form for subsequent splicing. Refer to the specific examples for reference.

[0017] If a color ultrasound image data sample is a 2 × 2 grayscale matrix, specifically:

[0018] Apply the vec operation to flatten it in row-major order to get Here H = 2, W = 2, M = 4, so

[0019] The vectorization of the device state parameters in this application includes: where, is the device state parameter vector of the i-th sample, belonging to P i is the original device parameter of the i-th sample, belonging to D p is the number of device parameters, because the device state parameter itself is a vector parameter, no conversion is needed. Refer to the specific examples for reference.

[0020] If the device parameters include gain, depth, and frequency, specifically: i = [0.8, 12.0, 3.5] T , gain = 0.8 dB, depth = 12.0 cm, frequency = 3.5 MHz, so Here

[0021] This application vectorizes the patient history label as follows: where, is the history label vector of the i-th sample, H i is the original history label of the i-th sample, which has been encoded as a real number vector, belonging to is the total dimension of the history feature, including numerical features and encoded category features. Refer to the specific examples for reference.

[0022] The history label includes: age, gender, diabetes history. The encoding rules are: gender: male encoded as 1, female encoded as 0; diabetes history: yes encoded as 1, no encoded as 0. Therefore, H i = [60, 1, 1] T , indicating age 60, male, and history of diabetes. So

[0023] The check type refers to the vectorization of the patient history label and the vectorization of the device state parameter. Finally, the constructed original data set includes a set of all composite feature vectors: Wherein, is the original data set, which is a set of N samples, x i is a composite feature vector, and N is the total number of samples. In order to facilitate the understanding of the technical scheme of the present application, a specific example is provided for reference.

[0024] There are two samples, N = 2, sample 1 T , the image in sample 2 Device parameter P2 = [0.9, 10.0, 4.0] T , history label: age = 45, female, diabetes history = no, The check type is a heart check, The composite vector x2 = [120, 180, 220, 70, 0.9, 10.0, 4.0, 0, 1, 0] T Therefore, the data set is

[0025] The further improvement of the technical scheme of the present application is that the scoring module is based on the color ultrasound image data I i as input, and the output is the SNR and MAR value of each sample, and the image quality is quantified according to the SNR and MAR value;

[0026] Wherein, the calculation of the SNR value includes: dividing the color ultrasound image into tissue region and background region, and obtaining the SNR value based on the logarithmic scale of the ratio of the signal power of the tissue region to the noise power of the background region;

[0027] The calculation of the MAR value includes: dividing the color ultrasound image into tissue region and background region, integrating the coordinate set of the motion artifact pixels in the tissue region, and obtaining the MAR value by comparing the total number of pixels in the image;

[0028] After the evaluation module obtains the SNR and MAR values, it is normalized to a unified dimension, and the comprehensive score Q is obtained by weighted fusion.

[0029] The further improvement of the technical scheme of the present application is that the scoring module is based on an organ contour segmentation algorithm to determine the target observation region of the color ultrasound image, such as the thyroid, heart valve, liver lesion area, as the tissue region, and other regions as the background region.

[0030] After obtaining the tissue region and the background region, a weight α is set according to the requirement of signal clarity and artifact stability in a clinical scene, and a comprehensive score Q is obtained based on the weight α, the SNR and the MAR values, that is, Q = 100 x (α·SNR + (1-α)·MAR), the score module automatically calculates based on the Otsu threshold method in each frame of the color ultrasound image data I i , avoids misjudgment of a fixed threshold for a dynamic sequence, and obtains the mean value of Q obtained from all frames of the color ultrasound image data I i according to the original data set as a score threshold, when Q≥score threshold, the score module outputs image quality good, and when Q i .

[0031] Further, wherein SNR i is the signal-to-noise ratio of the ith sample, P1 is the average signal power of the tissue region, P2 is the average noise power of the background region, wherein B is the pixel coordinate set of the background region, |B| is the number of background pixels, I i (x,y) is the pixel value of the image at position (x,y), μ B is the average of the background pixels. wherein F is the pixel coordinate set of the tissue region, |F| is the number of foreground pixels, B is defined as four corner pixels, that is, B = {(1, 1), (1, 3), (3, 1), (3, 3)}, and the pixel values are 5, 130, 15 and 20; therefore μ B =(5+130+15+20) / 4=42.5;

[0032] P2 = 1 / 4[(5-42.5) 2 +(130-42.5) 2 +(15-42.5) 2 +(20-42.5) 2 ]=2581.25;

[0033] In the tissue region, F is {(1, 2), (2, 1), (2, 2), (2, 3), (3, 2)}, and the pixel values are 120, 10, 150, 140 and 160.

[0034] Therefore, P1 = 16420-2581.25 = 13838.75, and therefore

[0035] Further, Wherein A is the coordinate set of motion artifact pixels, HxW is the total number of image pixels, the coordinate set of motion artifact pixels in the integrated organization area includes, A={(x,y)|G i (x,y)>τ.max(G i )}, wherein G i (x,y) is obtained by Sobel edge method detection, the coordinate set of motion artifact pixels, and τ is a proportional threshold, G i (x,y)>τ.max(G i ) represents that the coordinates of the motion artifact pixels screened by A are the detected abnormal gradient area, that is, the motion artifact pixels required by the application.

[0036] Further improvement of the technical scheme of the application is that, in the above, the color Doppler ultrasound image data in the original data set is scored by the scoring module, the preliminary screening and quality estimation of the color Doppler ultrasound image are completed, and then the device state parameters, patient history labels and examination types need to be screened.

[0037] The screening module takes the vectors converted from the patient history labels, examination types and device state parameters as input features, finds the screening threshold of the repeated data screening threshold through the neural network model from the historical data set, and screens the constructed original data set based on the obtained screening threshold to remove the repeated data.

[0038] Further improvement of the technical scheme of the application is that, in the above, the original data set is separated from the non-image features in the composite feature vector x i , a deduplication feature vector is constructed, and After the vectors of different dimensions are normalized, the cosine similarity between the historical data set z j and z i is calculated And a neural network model is constructed The optimal similarity threshold θ is learned from the historical data set, and the learned threshold θ is applied to deduplicate the current original data set.

[0039] Further improvement of the technical scheme of the application is that, in the screening module, the constructed neural network model The relationship between the optimal similarity threshold θ includes: Wherein θ is the screening value to be learned, that is, the optimal similarity threshold, is the historical data set and After obtaining the optimal similarity threshold θ, the average value of And the Sigmoid compression in the neural network model is 0 to 1.

[0040] Further, the similarity calculation function defines the similarity of the sample and the historical sample, where s(·,·) is a similarity score, · is a vector dot product operation, ||·|| is a Euclidean norm, are vectors converted from the history label, examination type, and device status parameter of the original data set and the historical data set respectively after normalization preprocessing, for example:

[0041] The dot product is 0.5*0.6+(-0.3)*(-0.2)+1.2*0.8=1.26, Finally

[0042] Based on the currently more mature neural network model to obtain the optimal similarity threshold θ, the initial original data set is At this time, the learned threshold θ is applied to the current data set to remove duplicates, and the determination is set to The screening rule is:

[0043] Traverse the samples in chronological order;

[0044] Keep sample i if and only if

[0045] If the current data set has three samples: (abdominal examination), (similarity 0.92 with 1), (similarity 0.75 with 1), θ=0.85, at this time, keep sample 1, remove sample 2 (0.92≥0.85), and keep sample 3 (0.75<0.85).

[0046] Further improvement of the technical scheme of the application is that after data feedback on the original data, it is ensured that the data source and the data itself have no problems, the composite vector is restored to an image matrix, and the color Doppler ultrasound image data are combined with the device status parameter, patient history label, and examination type for corresponding diagnosis and feedback.

[0047] Since there is more than one color Doppler ultrasound image device, and there is more than one group of color Doppler ultrasound image devices and original data sets, the execution module extracts and identifies the diagnostic features of the current diagnosis related anatomical structure and lesion area through the color Doppler ultrasound image target observation area features, specifically including:

[0048] The scoring module determines the color Doppler ultrasound image target observation area, and completes the diagnostic feature extraction of the color Doppler ultrasound image target observation area, including area, circularity, and average gray level.

[0049] In the above, the target observation region of the color Doppler ultrasound image was determined based on the organ contour segmentation algorithm. After obtaining the tissue region and background region, the tissue region was regarded as the ROI region in the existing image processing technology. Because in the discrete image space, the area calculation is based on pixel count, therefore, by ∑ (x,y)∈ROI I i Obtain the tissue area, and through To obtain the roundness, where The total area of ​​the color ultrasound image is determined by... Obtaining the average gray level and extracting diagnostic features from the target observation area of ​​the color ultrasound image can be based on extracting edge features by detecting areas with significant changes in brightness or color in the image. Commonly used operators include Sobel, Prewitt, Roberts, and Canny methods.

[0050] A further improvement of the technical solution of the present invention is that: the execution module encodes the examination type and medical history label in the original dataset into a demand vector r. i The driving screening strategy quantifies clinical needs into a computable vector, based on the demand vector r. i The quality score Q of the scoring module and the mean of diagnostic features are used to complete the strategy screening, which includes urgent, normal, and delayed.

[0051] Where S1 = Q·ω1 + (1-ω1)·C, where C is the mean of area, roundness and average gray level, Q is the quality score, and the strategy selection includes three intervals: [0.0.3], (0.3,0.6], and (0.6,1]. Therefore, the value of S1 determines the different intervals belonging to different strategies, where [0.0.3], (0.3,0.6], and (0.6,1] correspond to postponement, normal and emergency, respectively.

[0052] A further improvement of the technical solution of the present invention is that: after completing the strategy screening, the execution module schedules the display order according to the urgency level, and the color ultrasound image display device displays the color ultrasound image data, device status parameters, patient medical history labels, examination type and diagnostic features in the original dataset.

[0053] Compared with existing technologies, the beneficial effects of this invention are as follows: by constructing an original dataset and combining SNR and MAR, clinical scenario weights are introduced to improve image recognition accuracy. Moreover, compared with single grayscale evaluation, it can better meet the actual needs of different examination sites. Furthermore, a neural network model is used to learn dynamic thresholds from historical data. Compared with fixed thresholds, the redundancy data compression rate is improved, while the effective information retention rate is increased. In clinical practice, a demand vector is generated based on examination type and medical history labels to drive differentiated screening strategies. Through target area feature extraction and urgency ranking, the time for doctors to obtain key information is shortened. Attached Figure Description

[0054] Figure 1 a flowchart of the present application;

[0055] Figure 2 a system block diagram of the present application. DETAILED DESCRIPTION

[0056] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numbers in different drawings represent the same or similar elements. Although various aspects of embodiments are illustrated in the drawings, the drawings are not necessarily drawn to scale unless specifically noted.

[0057] The term "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0058] In addition, for the purpose of better illustrating the present application, numerous specific details are set forth in the following detailed description. Those skilled in the art will appreciate that the present application can be practiced without some or all of the specific details. In some instances, well-known methods, structures, and techniques have not been described in detail in order to avoid obscuring the present application.

[0059] The present application provides an overall scheme for realizing color ultrasound image processing control based on data feedback, comprising:

[0060] S10, collecting color ultrasound image data, device state parameters, and constructing an original data set based on structured patient medical history labels and examination types obtained from a hospital information system;

[0061] The device state parameters include gain, depth, and frequency, the patient medical history labels include numerical features and category features, and the examination types include category vectors.

[0062] The color ultrasound image data is composed of a 2D grayscale matrix, which is represented as a one-dimensional vector, the device state parameters are composed of a set of numerical parameters of gain, depth, and frequency, and are directly represented as a vector, the numerical features of the patient medical history labels include numerical components of age and disease duration, and the category features are gender and disease history, which are coded as numerical values to form a set of all the features described above, and the set is a composite feature vector, which is specifically:

[0063] The original color ultrasound matrix I representing the collected color ultrasound image data i ∈R H×W The flattened image vector is obtained by rearranging the matrix in row-major order into a column vector The device state parameter vector containing the original device parameters The medical history label vector containing the original medical history labels a check type vector comprising a sample check type establishing a composite feature vector The constructed original dataset comprises a set of all composite feature vectors.

[0064] Further, the application vectorizes the ultrasound image data, including: wherein, is the flattened image vector of the i-th sample, belonging to R M M is the dimension of the flattened vector, I i is the original ultrasound image matrix of the i-th sample, belonging to R H×W vec(·) is a matrix flattening operator, which rearranges the matrix in row-major order to a column vector, where M = H x W, H is the image height, W is the image width, and finally converts the image data to vector form for subsequent splicing. The present embodiment provides specific examples for reference.

[0065] If a sample of ultrasound image data is a 2 x 2 grayscale matrix, specifically:

[0066] Apply the vec operation to flatten in row-major order to obtain Here H = 2, W = 2, M = 4, so

[0067] The application vectorizes the device state parameters, including: wherein, is the device state parameter vector of the i-th sample, belonging to P i is the original device parameter of the i-th sample, belonging to D p is the number of device parameters, which in the present embodiment is gain, depth, and frequency, because the device state parameters themselves are vector parameters and do not need to be converted. The present embodiment provides specific examples for reference.

[0068] If the device parameters include gain, depth, and frequency, specifically: i P = [0.8, 12.0, 3.5] T gain = 0.8 dB, depth = 12.0 cm, frequency = 3.5 MHz, so Here D p = 3,

[0069] The application vectorizes the patient medical history label as: wherein, is the medical history label vector of the i-th sample, H iis the original medical history label of the i-th sample, which is encoded into a real number vector, belongs to is the total dimension of the medical history features, including numerical features and encoded category features, and the embodiment provides specific examples for reference.

[0070] The medical history label includes age, gender, and diabetes history. The coding rules are as follows: gender: male is coded as 1, and female is coded as 0; diabetes history: yes is coded as 1, and no is coded as 0. Therefore, H i =[60, 1, 1] T , which means the age is 60, the gender is male, and there is a history of diabetes. Therefore

[0071] The examination type refers to the vector representation of the patient's medical history label and the vectorization of the device state parameter. Finally, the original data set constructed includes a set of all composite feature vectors: wherein, is the original data set, which is a set containing N samples, x i is the composite feature vector, and N is the total number of samples. In order to facilitate the understanding of the technical solutions of the present application, the embodiment provides specific examples for reference.

[0072] There are two samples, N = 2, sample 1 is described in the above embodiment, x1 = [100, 150, 200, 50, 0.8, 12.0, 3.5, 60, 1, 1, 1, 0, 0] T , the image in sample 2 Device parameter P2 = [0.9, 10.0, 4.0] T , medical history label: age = 45, gender = female, diabetes history = no, examination type is cardiac examination, composite vector x2 = [120, 180, 220, 70, 0.9, 10.0, 4.0, 0, 1, 0] T , so the data set is

[0073] S20, based on the constructed original data set, quality score, based on SNR and MAR numerical quantification of image quality;

[0074] Based on the color ultrasound image data I i as input, the output is the SNR and MAR value of each sample, and the image quality is quantified according to the SNR and MAR value;

[0075] wherein, the calculation of SNR value includes: dividing the color ultrasound image into tissue region and background region, and obtaining the SNR value based on the logarithmic scale of the ratio of the signal power of the tissue region to the noise power of the background region;

[0076] MAR is worth calculating, including: dividing the ultrasound image into tissue area and background area, integrating the coordinate set of motion artifact pixels in the tissue area, and obtaining the MAR value with the ratio of the total number of image pixels;

[0077] After obtaining the SNR and MAR values, normalize to a unified dimension, and obtain the comprehensive score Q by weighted fusion.

[0078] Based on the organ contour segmentation algorithm, the target observation area of the ultrasound image is determined, such as the thyroid, heart valve, and liver lesion area, as the tissue area, and the other areas are as the background area;

[0079] After obtaining the tissue area and background area, set the weight α according to the requirement of signal clarity and artifact stability in the clinical scene, and the comprehensive score Q is based on the weight α, SNR and MAR values, and Q = 100 × (α · SNR + (1-α) · MAR) is obtained. Based on Otsu threshold method, the mean value of Q obtained from all frames of ultrasound image data I i is calculated automatically to avoid misjudgment of fixed threshold for dynamic sequence, and the mean value of Q obtained from all frames of ultrasound image data I i is taken as the score threshold, when Q≥score threshold, the output image quality is good, when Q<score threshold, the output image quality is poor, and the ultrasound image data I i is reacquired.

[0080] Further, wherein SNR i is the signal-to-noise ratio of the i-th sample, P1 is the average signal power of the tissue area, P2 is the average noise power of the background area, wherein B is the pixel coordinate set of the background area, |B| is the background pixel number, I i (x,y) is the pixel value of the image at position (x,y), μ B is the background pixel mean. wherein F is the pixel coordinate set of the tissue area, |F| is the foreground pixel number, in this embodiment, B is defined as four corner pixels, i.e. B = {(1,1), (1,3), (3,1), (3,3)}, and the pixel values are 5, 130, 15, and 20; therefore μ B =(5+130+15+20) / 4=42.5;

[0081] P2=1 / 4[(5-42.5) 2 +(130-42.5) 2 +(15-42.5) 2 +(20-42.5) 2 ]=2581.25;

[0082] The pixel values in the organization region F, i.e., {(1, 2), (2, 1), (2, 2), (2, 3), (3, 2)}, are 120, 10, 150, 140, and 160.

[0083] Therefore, P1 = 16420 - 2581.25 = 13838.75, and thus

[0084] Further, the motion artifact pixel set A is determined by the following formula: where A is a set of coordinates of motion artifact pixels, HxW is the total number of pixels in the image, and the set of coordinates of motion artifact pixels in the integrated organization region includes A = {(x, y) | G i (x, y) > t max(G i )}, where G i (x, y) is obtained by Sobel edge detection, is a set of coordinates of motion artifact pixels, and t is a proportional threshold, G i (x, y) > t max(G i ) represents that the coordinates of the motion artifact pixels selected by A are the detected abnormal gradient region, that is, the motion artifact pixels required by the present application.

[0085] S30, based on the neural network model, find the screening threshold of the repeated data screening threshold from the historical data set, and screen the constructed original data set based on the obtained screening threshold to remove repeated data;

[0086] In the above, the quality score of the color Doppler ultrasound image data in the original data set is obtained, the preliminary screening and quality estimation of the color Doppler ultrasound image are completed, and then the device state parameter, patient history label, and examination type need to be screened.

[0087] The vector converted from the patient history label, examination type, and device state parameter is used as an input feature, a neural network model is used to find a screening threshold of a repeated data screening threshold from a historical data set, and the constructed original data set is screened based on the obtained screening threshold to remove repeated data.

[0088] The non-image features in the composite feature vector x i of the original data set are separated to construct a deduplication feature vector, and x is obtained. After the vectors of different dimensions are normalized, the cosine similarity between the historical data set z j and z i is calculated. The optimal similarity threshold θ is learned from the historical data set, and the learned threshold θ is applied to deduplicate the current original data set.

[0089] The constructed neural network model The relationship with the optimal similarity threshold θ includes: Where θ is the screening value to be learned, that is, the optimal similarity threshold, is the historical data set and After obtaining the optimal similarity threshold θ, by And the mean of the Sigmoid compression in the neural network model to 0 to 1.

[0090] Further, the similarity calculation function defines the similarity of the sample and the historical sample, Where s(·,·) is the similarity score, · is the vector dot product operation, and ||·|| is the Euclidean norm, Respectively, the vectors converted by the preprocessed and respectively belonging to the original data set, the historical data set, the medical history label, the examination type, and the device state parameter, such as:

[0091] The dot product is 0.5*0.6+(-0.3)*(-0.2)+1.2*0.8=1.26, Finally

[0092] Based on the currently more mature neural network model to obtain the optimal similarity threshold θ, the initial original data set is At this time, the learned threshold θ is applied to the current data set to remove duplicates, and the setting is The screening rule is:

[0093] Traverse the samples in time sequence;

[0094] Keep sample i if and only if

[0095] If the current data set has three samples: (abdominal examination), (similarity 0.92 with 1), (similarity 0.75 with 1), θ=0.85, at this time, sample 1 is kept, sample 2 is removed (0.92≥0.85), and sample 3 is kept (0.75<0.85).

[0096] S40, based on the feature extraction of the target observation area of the color ultrasound image, the original data set is executed to perform the screening strategy, and the processed image data and context information are fed back to the color ultrasound image display device;

[0097] After the data feedback on the original data, it is ensured that the data source and the data itself have no problems, the composite vector is restored to the image matrix, and the ultrasound image data are combined with the equipment state parameters, patient history labels, and examination types for corresponding diagnosis and feedback.

[0098] Since there is more than one ultrasound image device, and there is more than one set of ultrasound image equipment and original data set, the current diagnosis-related anatomical structure and lesion area diagnosis features are identified through ultrasound image target observation area feature extraction, which specifically includes:

[0099] The ultrasound image target observation area is determined, and the ultrasound image target observation area completes the diagnosis feature extraction, including area, circularity, and average gray scale.

[0100] In the above, the ultrasound image target observation area is determined based on the organ contour segmentation algorithm, and after obtaining the tissue area and background area, the tissue area is regarded as the ROI area of image processing in the prior art. Because in the discrete image space, the area calculation is based on pixel counting, therefore, by ∑ (x,y)∈ROI I i The tissue area is obtained, and the average gray scale is obtained by The circularity is obtained, where is the overall area of the ultrasound image, and the average gray scale is obtained by The diagnosis feature extraction of the ultrasound image target observation area can be based on extracting edge features by detecting areas with obvious brightness or color changes in the image. Common operators include Sobel, Prewitt, Roberts, and Canny methods.

[0101] The examination type and history label in the original data set are encoded into a demand vector r i , which drives the screening strategy, and quantifies the clinical demand into a calculable vector based on the demand vector r i , the quality score Q, and the average of the diagnosis features. The strategy includes urgent, normal, and delayed.

[0102] Where S1=Q·ω1+(1-ω1)·C, C is the average of the area, circularity, and average gray scale, Q is the quality score, and the strategy screening includes [0.0.3], (0.3,0.6], and (0.6,1]. Therefore, the value of S1 determines the different intervals belonging to different strategies, where [0.0.3], (0.3,0.6], and (0.6,1] correspond to delayed, normal, and urgent, respectively.

[0103] After completing the strategy screening, the display order is scheduled according to the urgency, and the ultrasound image display device displays the ultrasound image data, equipment state parameters, patient history labels, examination types, and diagnosis features in the original data set.

[0104] In particular implementations, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium is capable of storing a computer program, and the computer program is executable to implement some or all steps of the application content of the data feedback-based color ultrasound image processing control system and each embodiment of the present application when executed by the data processing unit. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), or the like.

[0105] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present application can be realized by means of a computer program and its corresponding general hardware platform. Based on such understanding, the technical solutions in the embodiments of the present application can be embodied in the form of a computer program, i.e., a software product, which can be stored in a storage medium and includes a plurality of instructions for causing a device (which can be a personal computer, a server, a single-chip computer, an MCU, or a network device) containing a data processing unit to execute the method described in each embodiment or some parts of the embodiments of the present application.

[0106] The present application provides a data feedback-based color ultrasound image processing control system, and there are many methods and ways to realize the technical solutions. The above description is only the preferred embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, some improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application. The components not explicitly described in the embodiments can be realized by using existing technologies.

Claims

1. A color Doppler ultrasound image processing control system based on data feedback, characterized in that, The system includes: The acquisition and generation module is used to acquire color Doppler ultrasound image data and equipment status parameters, and to construct the original dataset based on the structured patient medical history tags and examination types obtained from the hospital information system. Scoring module: Quality scoring is performed based on the constructed original dataset, and image quality is numerically quantified based on SNR and MAR. The filtering module uses a neural network model to find a filtering threshold for duplicate data in historical datasets, and then filters the original dataset based on the obtained filtering threshold to remove duplicate data. The execution module extracts features from the target observation area of ​​the color ultrasound image, performs a filtering strategy on the original dataset, and feeds back the processed image data and contextual information to the color ultrasound image display device.

2. The color Doppler ultrasound image processing and control system based on data feedback according to claim 1, characterized in that, In the acquisition and generation module, the device status parameters include gain, depth, and frequency; the patient history tags include numerical features and categorical features; and the examination type includes a category vector. The color Doppler ultrasound image data consists of a 2D grayscale matrix, which is represented as a one-dimensional vector by the acquisition and generation module. The device status parameters consist of a set of numerical parameters for gain, depth, and frequency, which are directly represented as vectors by the acquisition and generation module. The numerical features of the patient's medical history tags include age and duration of illness, and the categorical features are gender and medical history. These categorical features are encoded as numerical values ​​by the acquisition and generation module. The set of acquisition and generation modules forms a composite feature vector, specifically: The acquisition and generation module will acquire the original color ultrasound matrix I, which represents the color ultrasound image data. i ∈R H×W By rearranging the matrix into column vectors in row-major order, the flattened image vector is obtained. And concatenate the device state parameter vector containing the original device parameters. Medical history tag vector containing original medical history tags Check type vector containing sample check types Establish composite feature vectors The original dataset constructed by the acquisition and generation module includes a set of all composite feature vectors.

3. The color Doppler ultrasound image processing and control system based on data feedback according to claim 1, characterized in that, The scoring module uses color Doppler ultrasound image data I i As input, the output is the SNR and MAR values ​​for each sample, and the image quality is quantified based on the SNR and MAR values. The calculation of SNR value includes: segmenting the color ultrasound image into tissue region and background region, and obtaining the SNR value based on the logarithmic scale of the ratio of signal power in tissue region to noise power in background region. The calculation of MAR includes: segmenting the color ultrasound image into tissue regions and background regions, integrating the coordinate set of motion artifact pixels in the tissue regions, and taking the ratio of this coordinate set to the total number of pixels in the image to obtain the MAR value; After obtaining the SNR and MAR values, the evaluation module normalizes them to a unified dimension and obtains the comprehensive score Q through weighted fusion.

4. The color Doppler ultrasound image processing and control system based on data feedback according to claim 3, characterized in that, The scoring module determines the target observation area of ​​the color Doppler ultrasound image based on the organ contour segmentation algorithm, which is used as the tissue area and other areas are used as the background area. After acquiring the tissue region and background region, a weight α is set according to the clinical scenario's requirements for signal clarity and artifact stability. The comprehensive score Q is based on the weight α, SNR, and MAR values, resulting in: Q = 100 × (α·SNR + (1-α)·MAR). The scoring module uses the Otsu thresholding method to evaluate each frame of color Doppler ultrasound image data. i The system automatically calculates and avoids misjudgment of dynamic sequences by fixed thresholds, based on all frames of color Doppler ultrasound image data in the original dataset. i The mean of the obtained Q values ​​is used as the scoring threshold. When Q ≥ the scoring threshold, the image quality output by the scoring module is excellent; when Q < the scoring threshold, the image quality output by the scoring module is poor, and the acquisition and generation module re-acquires color Doppler ultrasound image data I. i .

5. The color Doppler ultrasound image processing and control system based on data feedback according to claim 1, characterized in that, The filtering module takes vectors converted from patient medical history tags, examination types, and equipment status parameters as input features. Through a neural network model, it finds a filtering threshold for duplicate data in the historical dataset and filters the constructed original dataset based on the obtained filtering threshold to remove duplicate data.

6. The color Doppler ultrasound image processing and control system based on data feedback according to claim 5, characterized in that, The composite feature vector x of the original dataset i Separate non-image features from the image, construct a deduplicated feature vector, and obtain... After normalizing vectors of different dimensions, the historical dataset z is calculated using similarity. j With z i Cosine similarity between A neural network model f□ is constructed to learn the optimal similarity threshold θ from historical datasets, and the learned threshold θ is applied to remove duplicates from the current original dataset.

7. A color Doppler ultrasound image processing and control system based on data feedback according to claim 6, characterized in that, In the filtering module, the relationship between the constructed neural network model f□ and the optimal similarity threshold θ includes: Where θ is the selection value to be learned, i.e., the optimal similarity threshold. For historical datasets and After obtaining the optimal similarity threshold θ, through And the Sigmoid in the neural network model compressed to the mean of 0 to 1.

8. The color Doppler ultrasound image processing and control system based on data feedback according to claim 1, characterized in that, The execution module extracts features from the target observation area of ​​the color Doppler ultrasound image to identify diagnostic features of the anatomical structures and lesion areas relevant to the current diagnosis, specifically including: The scoring module determines the target observation area of ​​the color Doppler ultrasound image and extracts diagnostic features from the target observation area of ​​the color Doppler ultrasound image. The diagnostic features include area, roundness, and average gray level.

9. A color Doppler ultrasound image processing and control system based on data feedback according to claim 8, characterized in that, The execution module encodes the examination types and medical history labels in the original dataset into a demand vector r. i The driving screening strategy quantifies clinical needs into a computable vector, based on the demand vector r. i The quality score Q of the scoring module and the mean of diagnostic features are used to complete the strategy screening, which includes urgent, normal, and delayed.

10. A color Doppler ultrasound image processing and control system based on data feedback according to claim 9, characterized in that, After completing the strategy selection, the execution module schedules the display order according to the urgency level. The color ultrasound image display device displays color ultrasound image data, device status parameters, patient medical history labels, examination type, and diagnostic features from the original dataset.

Images